Neural Network

Project Description

This project is a modular neural network implementation in C++ designed for educational, experimental, and practical machine learning tasks. It provides core components for building, training, and evaluating neural networks, with a focus on extensibility and clarity. The codebase demonstrates how to implement and train a simple feedforward neural network for regression tasks, such as predicting housing prices.

Features

• Customizable feedforward neural network architecture
• Support for various activation functions (ReLU, Sigmoid)
• Mean Squared Error (MSE) loss function
• Stochastic Gradient Descent (SGD) optimizer
• Example demonstration with synthetic data
• Integration with Eigen for efficient matrix operations

Getting Started

Prerequisites

• C++20 compatible compiler (e.g., clang++, g++)
• Eigen library (install via Homebrew: brew install eigen)

Build and Run

1. Clone the repository:

   git clone <repo-url>
   cd neural_network

2. Build the demo:

   clang++ -g -std=c++20 src/main.cpp -I/opt/homebrew/include/eigen3 -o neural_network

3. Run the demo:

   ./neural_network

Project Structure

• src/ — Source code for neural network layers, activations, loss functions, optimizer, and main demo
  • NeuralNetwork.cpp — Main neural network class
  • LinearLayer.cpp — Linear layer implementation
  • ActivationFun.cpp — Activation functions (ReLU, Sigmoid)
  • LossFun.cpp — Loss functions (MSE)
  • Optimizer.cpp — Optimizer (SGD)
  • main.cpp — Demonstration and entry point
• include/ — Header files for all components
• data/ — Example datasets (CSV files)
• build/ — Build artifacts
• NN.so — Python bindings (if applicable)
• README.md — Project documentation

Example Demo

The demo in main.cpp generates synthetic data with a known linear relationship, trains the neural network, and prints predictions before and after training to show learning progress.